Air pollution is one of the main problems in the city of Bogotá, due to the high number of fixed and mobile sources found there. One of the pollutants that affect the health of the population are Sulfur Oxides (SOx). One way to reduce the risk associated with SOx is to create early alert systems that account for risk and have reliable forecasts. This research aims to evaluate the efficiency of a LongShort Term Memory (LSTM) network to predict the population risk produced by SOx. For this, design a LSTM model trained with Bogotá's air quality network. This model is able to predict 24h in advance and is evaluated for 15 of the highest polluted days in the period between 2013 and 2019 through statistical parameters (i.e., RMSE, r, IOA, MB) and criteria. The risk is calculated with the LSTM output for 5 different population groups and for every locality in the city. The results indicate that the model has very high precision for both daily and hourly values, and it is also found that that adult women and young men are the most affected by SOx, which means that policies need to take this into account, it is also found that the south-western polygon have the highest risk due to the number and type of mobile sources.